[dolphy]

Good day everyone!

I am building up workstation for simulations softwares. I am the beginner in this area and so this workstation will be used generally to learn simulation softwares. I am planning to run ANSYS, SolidWorks, COMSOL on this machine.

I am planning to spend as much as $2k. I also have pretty good discount with HP and Lenovo. It looks like that the workstation that they suggest for simulations has CPU Xeon Silver, but for the money I am planing to spend I only will be able to purchase a base line which is 1CPU 2.1 GHz - 8 cores. In the same price range I can purchase I-7 with much higher clock speed.

Which way would you suggest to go? Should I be looking at buying Xeon since it will give me some room for upgrade in the future (adding additional CPU, ECC RAM). Any suggestions will be appreciated.

Thanks

[flotus1]

Learning how to use the software you will mostly be doing single-threaded and lightly-threaded stuff. So an I5-8600k with 2x8GB of RAM and a small graphics card like the GTX 1050TI would be a pretty decent starting point. Way better than a Xeon with low clock speed that also requires 6 DIMMs to work best which drives up the price.
One step further would be an I7-7800x or 7820x with at least 4x8GB of RAM. This will allow you run some rather large simulation jobs as well. And it will outperform the Xeon workstation you mentioned in every single workload possible. Depending on where you buy, it might be difficult to fit this into your budget though.
Don't forget adding an SSD with at least 500GB, this will speed up tasks in pre- and post-processing which you will mostly be concerned with while learning how to use the software. And use DDR4-3000 or faster with both the I5 and I7. I really can not recommend a Xeon workstation for your purposes, not even with upgradeability in mind.

[dolphy]

Thank you Alex! According to another posts you are pretty knowledgable with this stuff.

I agree 100% with SSD memory.

I have some questions though:
- Why do you suggest DDR4-3000 or higher? On the intel web site they claim that memory type DDR4-2400 is used for I7-7800x, or maybe I missing something
- Do you know anything about Xeon W? It has a high clock speed around 3.6 - 4.0 GHZ and it will support EEC RAM as well, which seems to be a good way to for simulations.
- Will quadro video card benefit me with buying workstation in this price range?

Thanks a lot

[flotus1]

- Why do you suggest DDR4-3000 or higher? On the intel web site they claim that memory type DDR4-2400 is used for I7-7800x, or maybe I missing something

DDR4-2400 is the fastest memory speed Intel officially certifies for this processor. However, they can easily go way faster than that. For pedants this is overclocking. But it is a one-click procedure activating Intels own XMP and it involves very low risk of becoming unstable. Remember: memory speed matters in CFD.

- Do you know anything about Xeon W? It has a high clock speed around 3.6 - 4.0 GHZ and it will support EEC RAM as well, which seems to be a good way to for simulations.

Xeon W are the same chips as Skylake-X. Just more expensive with locked overclocking options but unlocked ECC capabilities. ECC is usually nothing you should be concerned about with your scope of application.

- Will quadro video card benefit me with buying workstation in this price range?

It will not benefit you, but it also would not hurt since you don't need much GPU performance. I would avoid spending more than 200$ on a GPU in your case.
Edit: well with the exception of Solidworks. Here a quadro card can be faster for some workloads than even the fastest GTX card https://www.pugetsystems.com/labs/ar...orks-2016-751/
A Quadro P600 would be a good idea for this application. The reason why I am usually tentative to recommend Quadro cards is that not many programs really make use of them and they are significantly more expensive than a Geforce card with similar raw performance.

[CFDBuddha]

It seems like i7-8700k is better for CFD than i7-7800x besides the fact that i7-8700k has just 2 memory channels. I tried to ask this question and elaborate it in thread: OpenFOAM benchmarks on various hardware, but for some reason I could not post there anything. Somebody else on i7-8700k vs i7-7800x?

[flotus1]

The differences between these two processors become pretty obvious in the thread you mentioned. You have to make a tradeoff between single-core performance (I7-8700k or rather I5-8600k) and parallel performance (I7-7800x). So I would disagree, I7-8700k is not the better CFD processor in general.

[CFDBuddha]

I alone was about to buy i7-7800x and my plan was to upgrade later (for 2-3 years) to i9-7940x or i9-7960x (or maybe upcoming Intel Cascade Lake-X) but then I saw a mentioned thread where it appears that i7-7820x is some kind of maximum for x299 platform (regarding CFD) - it performs the same as i9-7940x. So my plans for upgrade seems very limited and even impossible.

Then I saw some results of i7-8700k which seems better for CFD than i7-7800x - see attached pictures. On Euler 3D you can also see how much is i7-8700k faster than i5-8600k. Also, Simbelmyne posted on the mentioned thread results of the i7-8700k and Jeggi posted results for i7-5820k. Here again i7-8700k seems better than 4 memory channel i7-5820k which should not be much slower than i7-7800x.

Am I missing something?

[flotus1]

What's missing is that Euler CFD is not a good benchmark for real-world CFD applications. The results are all over the place, CPUs that are really old and slow (e.g. AMD FX) punch way above their league in this benchmark. Dual-channel CPUs beat quad-channel. SMT is extremely useful. High-core-count CPUs perform extremely well...it is all pretty counterintuitive. I would not read too much into these results. They have about as much relevance for CFD as cinebench scores.

The I7-5820k results that were posted do not reflect the potential of an I7-7800x. It ran at stock clock speeds and with slow memory. The results for the other Skylake-X CPUs with 6 cores yield a better estimate.
If I had to choose between the two platforms for CFD: Skylake-X with 6-8 cores, a slight overclock on mesh frequency and fast quad-channel memory.

[Jeggi]

The I7-5820k results that were posted do not reflect the potential of an I7-7800x. It ran at stock clock speeds and with slow memory.

I would also add that I used Windows! I will probably not install Linux anytime soon, but I can rerun the test with better RAM settings. Hopefully that will give some insight.

[CFDBuddha]

That would be nice!

[flotus1]

I once had a Xeon E5-1650v3. With CPU clocked at 4.4GHz and memory at DDR4-2666 instead of 2133 fluent parallel performance in double precision increased by more than 20%.

[CFDBuddha]

Then definitely i7-7800x should outperform i7-8700k even if i7-5820k with increased memory frequency equals to i7-8700k (based on your experience i7-5820k should also outperform i7-8700k).

[CFDBuddha]

I will go with i7-7800x. Question is what air cooler? Is Noctua NH-D15 enough to cool it below 75C when slightly overclocked?

Also, I understood the importance of memory frequency and regarding that what would be some lower limit of it? Maybe 2666MHz or even higher?

Thanks

[dolphy]

The best way to find out about the type of the RAM that intel suggest to use with their chip is to go to specs for the processor. I downloaded the app called Intel Arc for iphone. It has all the information about the processors and you can even compare them to see the difference. I would think that this is the minimum recommended. And as Alex mentioned if you are planing to overlock it (and most likely you will), look for higher MHz memory.

[flotus1]

I would think that this is the minimum recommended

Nope, it is the maximum supported frequency. This basically means that Intel does not guarantee trouble-free operation with higher frequencies. But the threshold value they set is VERY conservative with their current gen CPUs.

[CFDBuddha]

Memory prices are very high these days. Also there is a difference in prices between 2666 and 3200 MHz of course. If difference in performance between 2133 and 2666 is about 20% and for example if difference between 2666 and 3200 would be just 5% then 2666 is good to go (for me). I doubt that it goes linearly. That is why I asked for some lower limit of memory frequency (for CFD).

Intel ark specification is not that I asked for - it is just Intel general recommendation. I ask for some CFD recommendation which I think is higher than 2400.

Thanks.

[flotus1]

As soon as you hit the memory bandwidth limit, performance scales almost linearly with memory frequency. Which will be the case both for Coffee-Lake 6-cores and Skylake-X 8 cores. 6 cores not so much, but still quite noticeable.
And the price increase is not really steep, the only bright side of expensive memory in general. DDR4-3000 is not that much more expensive than DDR4-2400 nowadays.
If you can, get dual-rank modules. All 16GB UDIMM memory modules are still dual-rank. If you choose 8GB modules instead, choose wisely
https://geizhals.eu/?cat=ramddr3&xf=...M1%7E5831_DIMM

[CFDBuddha]

Thanks a lot. You were of great help flotus1!

[rickytrnr]

Hello

While those serious about simulation will mostly use a workstation with two Intel Xeon CPUs

___________________________________________
Best site alertfreejob in India

[lucabiagioni]

Hi everyone,

I'm in the same situation and I found this cheap workstation for 700$:

INTEL XEON QUAD CORE E5-1620 V2 3,7Ghz (Turbo Max 3,9Ghz) 10MB smart CACHE
RAM 16GB DDR3
HARD DISK 600GB SAS (2x300gb)
NVIDIA QUADRO K2000 2GB DDR5


What do you think about it?